AI Paper Atlas

Research map / Vision-language and multimodal / CLIP and zero-shot

PACTran: PAC-Bayesian Metrics for Estimating the Transferability of Pretrained Models to Classification Tasks

Nan Ding, Xi Chen, Tomer Levinboim, Soravit Changpinyo, Radu Soricut

ECCV 2022 · Vision-language and multimodal › CLIP and zero-shot · approach: Parameter-efficient visual tuning · 28 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
use PT-Gauss grid instead of LEEPtransferability metric · NeuCRaB, N/K=2 avgKendall-Tau correlation: 0.083 -> 0.680Improved
use GMM assignments (N-LEEP) instead of LEEPLEEP · NeuCRaB, N/K=2 avgKendall-Tau correlation: 0.083 -> 0.401Improved
use PT-Gauss fix instead of LINEAR-VALIDtransferability metric · NeuCRaB, N/K=10 avgKendall-Tau correlation: 0.480 -> 0.641Improved
add flatness regularizer to linear classifier loss (PT-Gauss fix vs LINEAR) (ablation)LINEAR · NeuCRaB, N/K=2 avgKendall-Tau correlation: 0.139 -> 0.654Improved
use PT-Gauss fix instead of PT-Gauss grid when K is smallPT-Gauss · NeuCRaB, N/K=5 avgKendall-Tau correlation: 0.641 -> 0.673Improved
use PT-Gauss grid instead of fix at N/K=2PT-Gauss · NeuCRaB, N/K=2 avgKendall-Tau correlation: 0.654 -> 0.680Improved
use PACTran-Dirichlet/Gamma with GMM assignments instead of N-LEEPN-LEEP · NeuCRaB, N/K=10 avgKendall-Tau correlation: 0.497 -> 0.477No clear effect
use LINEAR-VALID instead of LINEAR for small KLINEAR · NeuCRaB, 2-9 classes, N/K=10Kendall-Tau correlation: 0.109 -> 0.123Mixed
use PT-Gam instead of LEEPLEEP · OKVQA, N=200 TotalKendall-Tau correlation: 0.508 -> 0.532Improved
use PT-Gauss grid instead of LINEAR-VALIDLINEAR-VALID · OKVQA, N=200 TotalKendall-Tau correlation: 0.528 -> 0.582Improved
use PT-Gauss fix instead of penultimate feature extraction costfeature extraction · NeuCRaB, N/K=10, 100+ classesGFLOPS: 3.88E3 -> 6.45E1Improved
balance flatness and empirical-risk terms (ratio about 1.0) (ablation)PT-Gauss · NeuCRaB VTABKendall-Tau correlationImproved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy